Handling timestamps in .h5 files with rhdf5

[glenncampagna]

• Summary of last week's findings about the timestamp issue in R
• We used the H5Dread command and utilized the bit64conversion argument in order to get usable values, since R does not support the raw 64-bit timestamps

  fid = H5Fopen("OR1_7700_7980.h5")
  did = H5Dopen(fid, "RESULTS/RCHRES_R001/HYDR/table")
  rchres1 <- H5Dread(did, bit64conversion= "double")
  head(rchres1)
       index       DEP     IVOL O1 O2       O3 OVOL1 OVOL2     OVOL3 PRSUPY
  1 4.417668e+17 0.2415072 8.847264  0  0 2.055183     0     0 0.1229398      0
  2 4.417704e+17 0.3002159 8.828485  0  0 3.175832     0     0 0.2161576      0
  3 4.417740e+17 0.3486096 8.810900  0  0 4.282218     0     0 0.3081839      0
  4 4.417776e+17 0.3905506 8.793962  0  0 5.374578     0     0 0.3990412      0
  5 4.417812e+17 0.4279465 8.777404  0  0 6.453111     0     0 0.4887475      0
  6 4.417848e+17 0.4619085 8.761090  0  0 7.517995     0     0 0.5773184      0
      RO     ROVOL     SAREA        TAU     USTAR      VOL VOLEV
  1 2.055183 0.1229398  67.47366 0.02344255 0.1099862 15.79433     0
  2 3.175832 0.2161576  83.87602 0.02914127 0.1226281 24.40666     0
  3 4.282218 0.3081839  97.39652 0.03383873 0.1321425 32.90938     0
  4 5.374578 0.3990412 109.11423 0.03790982 0.1398658 41.30430     0
  5 6.453111 0.4887475 119.56212 0.04153978 0.1464090 49.59296     0
  6 7.517995 0.5773184 129.05061 0.04483640 0.1521076 57.77673     0

• Now converting the timestamp to a date-time

  origin <- "1970-01-01"
  rchres1$index <- as.POSIXct((rchres1$index)/1000000000, origin = origin, tz="UTC")
  > head(rchres1)
              index       DEP     IVOL O1 O2       O3 OVOL1 OVOL2     OVOL3
  1 1984-01-01 01:00:00 0.2415072 8.847264  0  0 2.055183     0     0 0.1229398
  2 1984-01-01 02:00:00 0.3002159 8.828485  0  0 3.175832     0     0 0.2161576
  3 1984-01-01 03:00:00 0.3486096 8.810900  0  0 4.282218     0     0 0.3081839
  4 1984-01-01 04:00:00 0.3905506 8.793962  0  0 5.374578     0     0 0.3990412
  5 1984-01-01 05:00:00 0.4279465 8.777404  0  0 6.453111     0     0 0.4887475
  6 1984-01-01 06:00:00 0.4619085 8.761090  0  0 7.517995     0     0 0.5773184
  PRSUPY       RO     ROVOL     SAREA        TAU     USTAR      VOL VOLEV
  1      0 2.055183 0.1229398  67.47366 0.02344255 0.1099862 15.79433     0
  2      0 3.175832 0.2161576  83.87602 0.02914127 0.1226281 24.40666     0
  3      0 4.282218 0.3081839  97.39652 0.03383873 0.1321425 32.90938     0
  4      0 5.374578 0.3990412 109.11423 0.03790982 0.1398658 41.30430     0
  5      0 6.453111 0.4887475 119.56212 0.04153978 0.1464090 49.59296     0
  6      0 7.517995 0.5773184 129.05061 0.04483640 0.1521076 57.77673     0

  The same process should work for converting the timestamps of any data table in the .h5 file to a date-time

[glenncampagna]

• Writing a dataset from the .h5 to a .csv:
• Make sure the output of H5Dread has been assigned to a variable and is in the format of a data frame, mine is 'rchres1'
• write.table(rchres1, file = "/Users/glenncampagna/Documents/HARPteam22/rchres.csv", sep = ",", row.names = FALSE) results in a populated .csv file in my HARPteam22 folder
• Note: this was only successful in the local R application, and would need a different file path to be downloaded while on the deq1 server

[rburghol]

Ok so the discovery of bit64conversion is exciting -- great work! Now if we can just figure out why we have to divide by a billion.

[nicoledarling]

@rburghol I believe since the Unix timestamps are in nanoseconds and the normal dates are in seconds, we need to divide by a billion to convert. More on this in link below in answer 5: https://discuss.dizzycoding.com/convert-numpy-datetime64-to-string-object-in-python

[juliabruneau]

@rburghol in addition to what @nicoledarling found, this converter website gives the definition of the Unix epoch timeseries that the .h5 files use:

The Unix epoch (or Unix time or POSIX time or Unix timestamp) is the number of seconds that have elapsed since January 1, 1970 (midnight UTC/GMT), not counting leap seconds (in ISO 8601: 1970-01-01T00:00:00Z). Literally speaking the epoch is Unix time 0 (midnight 1/1/1970), but 'epoch' is often used as a synonym for Unix time. Some systems store epoch dates as a signed 32-bit integer, which might cause problems on January 19, 2038 (known as the Year 2038 problem or Y2038). The converter on this page converts timestamps in seconds (10-digit), milliseconds (13-digit) and microseconds (16-digit) to readable dates.

Since they state that "some systems store the epoch dates as signed 32-bit integer", this might be the reason why we had to convert the data first with bit64conversion?

From: https://www.epochconverter.com/

[rburghol]

@nicoledarling ahh this is a good catch. I would clarify though that unix tjmestamps are, i think in seconds, and it is numpy timestamps that are in nanoseconds. numpy is a library in python, and perhaps there is some reason that it is storing these data as nanoseconds. Maybe the way our ches bay UCI files are formatted? I think we should look at the output of the test UCIs/h5 files that come with the hsp2 package.
Download and run these thru hsp2:

• https://github.com/respec/HSPsquared/raw/master/tests/test10/HSP2results/test10.wdm
• https://raw.githubusercontent.com/respec/HSPsquared/master/tests/test10/HSP2results/test10.uci